Optimal Search for Product Informationpeople.bu.edu/monic/search_slides.pdf · 2018. 3. 5. ·...
Transcript of Optimal Search for Product Informationpeople.bu.edu/monic/search_slides.pdf · 2018. 3. 5. ·...
Optimal Search for Product Information
Fernando Branco Monic Sun J. Miguel Villas-BoasUniversidade Catolica Portuguesa Stanford University Berkeley-Haas
A Question to the Consumer
Monthly time spent on search: 138 mm hours
Are you using more or fewer sites when doing
product research online compared to last year?
(a study done by ExpoTV.com)
Diverse responses
I use just as many sites as often as I did last year..
Definitely more.
…I actually use fewer sites than I used to for product
research.
How informative, easiness
2
What Happens During Information Search?
t
3
Expected
Value
Research Questions
When should the consumer stop searching for
more information?
How does search informativeness matter?
How does search cost matter?
Does the seller benefit from more or less
consumer search?
What is the seller’s optimal pricing strategy?
What is his optimal information provision strategy?
4
The Model
One consumer, one product, one seller
The consumer learns some news on an aspect of “product fit” in each step of search, His “true” utility given the T product attributes is
where equals z or –z with equal probability
z can be different across attributes, is “news” when checking attribute i
ix
T
i
ixvU1
5
ix
After inspecting t attributes, the consumer’s expected utility is
As z goes to zero and T goes to infinity (an infinite mass of attributes), the process becomes a Brownian motion:
Expected Utility through the Search Process
T
ti
i
t
i
i xExvu11
)(
ddu
6
The Consumer’s Problem
At each point of time, consumer has to
optimally choose among
1. Continue to gather more information at cost c
per (unit of) attribute searched
2. Stop searching, buy the product
3. Stop searching, without buying the product
Expected utility if keep on searching:
),(),( dttduuEVcdttuV
7
Getting V(u)
Taylor Expansion (plus Ito’s lemma):
Equals Zero
8
Boundary Conditions
Intuition:
suppose V΄(Ū)ˉ<1, then it would pay off to search more once reaching Ū a contradiction.
Suppose V΄(Ū)ˉ>1, then it would pay off to stop search prior to reaching Ū a contradiction.
High contact condition
9
The Value Function
u
c16
2
c4
2
V
cU
4
2cU
4
2 0
No
purchase
Purchase
Keep
searching
45°
cuu
cuV
162
1)(
22
2
10
The Optimal Stopping Rule
The two bounds are
symmetric around 0
Starting point v
does not affect the
boundaries
Purchase threshold
increases with σ
and decreases with
c
11
Purchase Likelihood I
Ū
U
v
t0
u(t)
Utu )( 1
)( 1tuU
1t)( 1tu
12
Purchase Likelihood II
Formally,
Prior to any search,
If v<0, having each attribute be more important increases the purchase likelihood (greater possibility of changing preferences)
If v<0, lower search cost also leads to a greaterpurchase likelihood (cheaper to gain information to reverse preferences)
Results change if v>0.
UU
Uuu
)Pr(
2
12)Pr(
2
cvv
13
Optimal Price
Changing the price essentially changes the starting
valuation, and hence changes the purchase
likelihood linear demand (marginal cost is g)
No
search
Price if set after
search
Price if search is
impossible
14
Maximum profit (in expectation) is
always increases with v
increases with Ū if v<g+ Ū
Low v: increase in price dominates
High v: decrease in purchase likelihood dominates
Consumer surplus is half of the optimal profit: does not always increase with informativeness and
decrease with search cost
Profit
15
No
search
Extension 1: Independent Signals
t decreases in t at a decreasing rate. V(u,t) and purchase and no-purchase
thresholds depend on the number of signals t already checked
16
Extension 2: Finite Mass of Attributes
When consumer is close to checking all possible attributes, it is not possible to
raise expected value of the product substantially
17
Other Extensions
3. Discounting: If positive expected value of purchase, keeping on
searching is more costly (more likely to purchase the products)
Conclusion: purchase threshold is closer to zero than exit threshold.
4. Choosing the search intensity: when consumer is closer to the purchase
threshold, he searches more intensely, as discounting makes it more costly
to keep on searching.
Conclusion: not to search intensively if far away from purchase, and search
intensively when close to purchase.
18
Conclusions
Parsimonious model of search for information
Stopping rule obtained optimally as a function of search costs and information gained
Implications for pricing – pricing affects consumer search behavior
Extensions to signals for value of a product, finite mass of attributes, discounting, intensity of search
Other questions:
Implications for social welfare: more search more correct choices
Search over multiple alternatives (different from Gittins index problem)
Optimal provision of information if different attributes provide different amount of information
19
Thank You!